International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.

169-174
Geotec., Const. Mat. & Env., DOI: https://doi.org/10.21660/2018.48.99305
ISSN: 2186-2982 (Print), 2186-2990 (Online), Japan

COMPRESSIVE STRENGTH MODELLING OF CONCRETE MIXED

WITH FLY ASH AND WASTE CERAMICS USING k-NEAREST
NEIGHBOR ALGORITHM

*Kenneth Jae T. Elevado1, Joenel G. Galupino1 and Ronaldo S. Gallardo1

1
Civil Engineering Department, De La Salle University, Manila, Philippines

*Corresponding Author, Received: 26 March 2018, Revised: 8 April 2018, Accepted: 26 April 2018

ABSTRACT: Excessive materials are being manufactured, and along with it are the waste products that are
being produced due to the rapid growth of industries. In the Philippines, wastes such as fly ash and damaged
ceramics are being considered as a construction material since there are recent researches that proved their
properties are comparable to cement and aggregates. In this study, compressive strength tests (ASTM C 39)
were conducted to obtain the compressive strength of the concrete mixed with varying amounts fly ash and
waste ceramics. Moreover, specimens were also subjected to varying days of curing to assess their strength
development. Due to the availability of a wide range of data, machine learning model, such as the k-nearest
neighbor, were also considered; it can predict an unknown target parameter without consuming tremendous
time and resources. Thus, this study aimed to provide a k-nearest neighbor model that will serve as a
reference to predict the compressive strength of concrete while incorporating waste ceramic tiles as a
replacement to coarse aggregates while varying the amount of fly ash as a partial substitute to cement. The k-
nearest neighbor model used was validated to ensure the predictions are acceptable.

Keywords: Compressive strength, waste utilization, fly ash, ceramics, nearest neighbor

1. INTRODUCTION for measuring the similarity of a set of objects

Waste utilization has been one of the
considerations in the construction industry towards
sustainability. With the growing demand in the
construction industry, excessive materials are
being manufactured, and along with it, are the
waste products that are being produced. In the
Philippine context, there is an existing problem
with regards to fly ash since the country mostly
relies on coal-based electricity generation. There is
also no ash industry in the Philippines that exists to
consume all the fly ash wastes produced, which
amounts to approximately 300,000 tons per annum
[1]. As a result, the fly ash utilization in the
country is underdeveloped. Moreover, most of the
construction and demolition wastes worldwide are
composed of ceramic materials [2] and these waste
materials are only disposed of in landfills.
To address sustainability, these wastes, such
as damaged ceramic tiles and fly ash, are being
considered as a construction material since there
are recent researches that proved their properties
are comparable to cement and aggregates [3-8].
Due to the availability of a wide range of data
gathered during the experiment, machine learning
model, such as the k-nearest neighbor, were also
considered; it can predict an unknown target
parameter without consuming tremendous time
and resources. The k-nearest neighbor (k-NN)
algorithm was used because it is mainly employed
based on some measures of distance and is one the
oldest pattern classifier methods with no required
pre-processing [9]. Thus, this study aimed to
provide a k-nearest neighbor model that will serve
as a reference to predict the compressive strength
of concrete while incorporating waste ceramic tiles
as a replacement to coarse aggregates while
varying the amount of fly ash as a partial substitute
to cement.
2. METHODOLOGY
Mix design of the specimens was prepared
referring to the American Concrete Institute
standards [10]. Moisture content, specific gravity,
and absorption tests [11-12] of the components of
concrete and as well as its unit weight and voids
[13] were the prerequisites in preparing for the mix
design. The summary of the results is shown in
Table 1. In this study, the tests are similar to
previous studies [14-15] for the formulation of the
mix design of the study. Similarly, chemical
analysis of fly ash used in the experiment was
conducted [16].
This study incorporated the use of waste
ceramic tiles in replacing gravel as the coarse
aggregates in the design mix with the following
substitutions: 0%, 18.25%, 37.5%, 56.25% and
75%. The percentages of replacements of ceramic
tiles were patterned after the previous studies [8,

169
 International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.169-174

17]. On the other hand, Type 1 Portland Cement M14 F12.5 C56.25 12.50% 56.25%
was also replaced by fly ash in this study M15 F50 C75 50.00% 75.00%
M16 F0 C75 0.00% 75.00%
considering five replacements: 0%, 12.5%, 25%, M17 F25 C75 25.00% 75.00%
37.5% and 50%. All replacements of both waste
ceramic tiles and of fly ash were in terms of mass Washed sand and river gravel were used as the
percentage. The control mix had pure cement and conventional fine and coarse aggregates of
gravel. Considering the replacements of waste concrete, respectively while substituting the later
ceramic tiles and fly ash and the output from the with glazed ceramic tiles. Although replacements
Design of Experiments (DOE) conducted, a total were considered in the coarse aggregates of
of 17 mixes were prepared in this study. concrete, it was ensured that the sizes in each mix
still adhere to the standards through performing
Table 1. Summary of material properties sieve analysis [18] in all mixes. The waste ceramic
Description Results tiles were manually crushed, if needed, and sieved
The dry rodded density of gravel 1567.839 kg/m3 to ensure that the grain size distribution follows
Specific Gravity of Cement 3.150 the standard stipulations [18].
Specific Gravity of gravel 2.812 Considering 25-100 mm slump and a
maximum size of 19.0 mm of the aggregates, the
Specific Gravity of sand 2.505
estimated mixing water that should be used is 184
Moisture Content of gravel 0.349%
kilogram per cubic meter of concrete. In addition,
Moisture Content of sand 1.566% no chemical admixture was used in the study.
Absorption of gravel 1.639% Furthermore, the water-cement ratio that was
Absorption of sand 2.765% considered in this study upon further interpolation
Fineness modulus of Sand 2.760 is 0.478. It was derived by having a target nominal
Chemical Compounds of Fly Ash
compressive strength of 28 MPa, which is
typically used in the industry. This water-cement
Silicon Dioxide (SiO 2 ) 49.6%
ratio was kept constant for all mixes. A total of
Aluminum Trioxide (Al 2 O 3 ) 26.7% 306 specimens were prepared to accommodate the
Ferric Oxide (Fe 2 O 3 ) 4.26% curing at 3 ages: 7, 28 and 56 days.
Calcium Oxide (CaO) 8.2% Moreover, compressive tests were performed
Magnesium Oxide (MgO) 5.9% after the specified day of curing to determine the
Sulfur Trioxide (SO 3 ) 0.83%
physical properties of the specimens. The load was
applied to the specimen, and the maximum load
that the specimen could carry was recorded. With
In order to have a systematized way of this, the compressive strength was computed by
labeling the specimens, Mix IDs were used in the simply dividing the maximum compressive load
study. The acronyms “F” and “C” refer to fly ash that the specimen was able to carry by its average
and waste ceramic tiles, respectively. The number cross-sectional area.
that immediately follows the acronym signifies it Once the data are available k-NN modeling
percentage replacement to either cement or gravel.
commenced. Each k-NN model consists of a data
For example, the mix ID “F50C75” refers to the case having a set of independent variables labeled
mix with 50% fly ash, 50% Type 1 Portland by a set of dependent outcomes, the research k-NN
Cement, 75% waste ceramic tiles and 25% gravel. model classification of is shown in Fig. 1. The
Mix IDs are shown in Table 2: independent and dependent variables can be either
continuous or categorical. In the study the
Table 2. Mix IDs of the specimen
dependent and independent variables are shown in
Ceramic
Fly Ash Table 3:
Mix No. Mix ID Tiles
Content
Content
M1 F0 C0 0.00% 0.00% Table 3. Dependent and Independent Variables
M2 F50 C0 50.00% 0.00% Independent
M3 F50 C 37.5 50.00% 37.50% Dependent Variable(s)
Variable(s)
M4 F25 C37.5 25.00% 37.50%
1. 7-Days Compressive Strength
F37.5 C 1. Fly Ash %
M5 37.50% 18.75% 2. 28-Days Compressive Strength
18.75 2. Ceramics %
3. 56-Days Compressive Strength
M6 F25 C0 25.00% 0.00%
M7 F25 C 18.75 25.00% 18.75%
M8 F12.5 C 37.5 12.50% 37.50% The k-nearest neighbor model used was
M9 F37.5 C37.5 37.50% 37.50%
M10 F0 C37.5 0.00% 37.50%
validated to ensure the predictions are acceptable.
M11 F37.5 C56.25 37.50% 56.25%
F12.5 C
M12 12.50% 18.75%
18.75
M13 F25 C56.25 25.00% 56.25%

170
 International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.169-174

Compressive strength tests were conducted at

Fig. 1. k-NN model classification for the study
3. RESULTS & DISCUSSION
three curing periods: 7 days, 28 days and 56 days.
A sample plot is shown in Fig. 2. This was done in
order to have a representation of the compressive
strengths of all mixes at early, nominal and late
stages for further analysis. The results are shown
in Fig. 3.
It could be inferred that M17 or F25C75
consistently had the highest compressive strengths
among all the other mixes at all stages. On the
other hand, M2 or F50C0 had the least
compressive strengths at all periods. However, it
could be seen that at the 56-day period, the
compressive strength of F50C0 is very close to but
not greater than the strength of the conventional
mix. This tremendous increase in the strength of
F50C0 from its 28-day strength is due to the
pozzolanic reaction.

3.1 Compressive Strength

The compressive strengths of the conventional

mix at 7, 28 and 56 days of curing periods were
21.645 MPa, 28.302 MPa, and 28.722 MPa,
respectively. This implies that the 28-MPa target
nominal strength was achieved. Moreover, F50C0
(M2) and F25C75 (M17) consistently achieved the
minimum and maximum compressive strengths
among all mixes at all ages. The results are shown
in Table 4:

Table 4. Compressive strengths of all mixed

Compressive Strengths Fig. 2. 7-Days Compressive Strength (Early)
MIX (MPa)
MIX ID
No.
7-day 28-day 56-day
M1 F0 C0 21.65 28.30 28.72
M2 F50 C0 19.07 26.34 28.41
M3 F50 C 37.5 21.68 27.85 33.88
M4 F25 C37.5 27.29 37.21 39.20
M5 F37.5 C 18.75 24.97 32.14 35.44
M6 F25 C0 25.51 33.88 38.09
M7 F25 C 18.75 27.78 34.51 38.95
M8 F12.5 C 37.5 30.96 37.44 41.57
M9 F37.5 C37.5 25.07 33.83 35.97
M10 F0 C37.5 29.00 32.14 38.00
M11 F37.5 C56.25 27.46 34.49 38.32
M12 F12.5 C 18.75 26.00 34.06 37.90
Fig. 3. Strength development of the mixtures
M13 F25 C56.25 25.15 34.17 38.28
M14 F12.5 C56.25 27.53 27.23 36.36
All compressive strengths at the 56-day period
M15 F50 C75 24.53 33.21 41.78 exceeded the strength of the conventional mix.
M16 F0 C75 24.86 33.95 39.66 Similarly, it is due to the pozzolanic reaction
M17 F25 C75 32.06 38.11 44.70 brought by fly ash and the waste ceramic tiles as
Minimum 19.07 26.34 28.41 well. Furthermore, it could be observed from the
Maximum 32.06 38.11 44.70 same Fig. that there is a slight difference between
the projection of the strength development of M14

171
 International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.169-174

or F12.5C56.25 and of the conventional mix (M1 respectively, while maximizing the compressive
or F0C0). strength (labeled “f’c”). Table 5 shows the
The conventional mix attained 30.76%, 32.7% summary of the optimization constraints
and 1.48% increase in the compressive strength considered:
from 7 to 28, 7 to 56 and 28 to 56 days of curing
periods, respectively. The 1.45% increase from 28 Table 5. 28-Days Strength optimization constraints
to 56 days was observed to be the least percentage Lower Upper
Name Goal
increase among all mixes; next to it was the Limit Limit
compressive strength of F25C37.5 with only %F is in range 0.00 0.50
5.36% increase. In terms of the least percentage %C is in range 0.00 0.75
increase at 7 to 28 and 7 to 56 days, F12.5C56.25 f'c maximize 26.34 38.11
and F0C37.5 provided the least increase with only
1.10% and 31.03%, respectively.
With the optimization, it could predict the
The maximum percentage increase at 7 to 28, 7 resulting compressive strength given the
to 56 and 28 to 56 days of curing periods were percentage replacements of fly ash and waste
observed at F50C0, F50C75 and F12.5C56.25 with ceramic tiles. With 0.92 desirabilities, Equation 1
38.12%, 70.32%, and 33.55%, respectively. The was the generated.
tremendous increase of F50C75 from 7 to 56 days
was brought about by the pozzolanic reaction since f’c = 30.01565 + 33.35915 %F +
it was established that both waste materials used, 0.98742 %C + 8.83664 %F %C – (Eq. 1)
fly ash and glazed ceramic tiles, possessed 79.86914 (%F)2 + 2.53592 (%C)2
pozzolanic properties. Where:
A sample of the pozzolanic reaction is shown in f’c = predicted strength (MPa);
Fig.s 4 and 5 where Scanning Electron Microscopy %F = percentage of fly ash;
(SEM) images were captured at 7 and 28 days of %C = percentage of waste ceramic tiles.
curing periods. As seen in the Fig.s, less inter-
particle voids were present at the SEM image of The effects of cement and coarse aggregates
the latter. This suggests that fibrillation occurred as modification as previously discussed were also
the curing period was increased, which is an effect observed –an increase in %FA resulted to an
of the pozzolanic properties of both fly ash and increase in strength up to an optimum amount and
waste ceramic tiles (WCT). an increase in %WCT yielded an increase in
strength. The combination of both waste materials
was gradually producing higher compressive
strengths.

3.2 k-Nearest Neighbor Algorithm Model

k-Nearest Neighbor (k-NN) algorithm is one

Fig. 4. SEM images of the F25C75 (M17) at 7-
of the simplest classification algorithms and it is
day curing period (from left to right: 500X, 100X
one of the most used learning algorithms [19-20].
and 1500X magnifications)
k-NN predictions are based on the intuitive
assumption that objects close in distance are
potentially similar, it makes good sense to
discriminate between the K nearest neighbors
when making predictions [21].
In the k-Nearest Neighbor (k-NN) algorithm,
Fig. 5. SEM images of the F25C75 (M17) at 28- the following were considered, shown in Table 6:
day curing period (from left to right: 500X, 100X
Table 6. Considerations in the k-Nearest Neighbor
and 1500X magnifications)
(k-NN) algorithm
The 28-Days compressive strength as optimized. Description Value
The percentage replacements of fly ash and waste Sample Size 75%
ceramic tiles (labeled “%F” and “%C”, Testing Size 25%
respectively) were accounted along with its v-Value 10
corresponding treated data. In terms of the Seed 1000
optimization constraints, %F and %C were
maintained to be in range at 0 to 0.5 and 0 to 0.75,

172
 International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.169-174
The resulting k-optimal is 2, which means that
there are 2 closest training samples were selected
based on a distance metric and voted for the most
number of samples per class.
To validate, the Observed Compressive
Strength vs. Observed Compressive Strength by
the k-Nearest Neighbor (k-NN) algorithm were
compared. A line that shows equality between the
variable observed (Experimental Data) on the
horizontal axis of a diagram and the variable
predicted (k-Nearest Neighbor (k-NN) algorithm)
on the vertical axis. The plots should be near the
equality line to ensure the predictions are
acceptable, shown in Fig. 6.
Fig. 6. Equality line of the k-Nearest Neighbor
(k-NN) algorithm Model
As seen in Fig. 6, it could be said that although
there were some data points far from the equality
line, most of the data points were still close to the
projected line. This means that there was a small
residual observed between the experimental data
and the theoretical or projected data.
4. CONCLUSIONS &
RECOMMENDATIONS
Based on the test results, the conventional mix
attained its target nominal strength with 28.302
MPa. Among all modified mixes, F50C0 and
F25C75 resulted in the least and highest
compressive strengths at all ages with 26.343-MPa
and 38.112-MPa nominal strengths, respectively.
All mixes had an increasing nominal compressive
strength when waste ceramic tiles replacement was
also increased except for the mix with 12.5% fly
ash replacement, where the strength decreased
from 37.5% to 56.25% waste ceramic tiles
substitution. In terms of cement variation, all
combinations showed an increasing nominal
strength up to an optimum amount. Based from the
173
experimental data, all combinations have shown an
optimum amount of 20% to 30% fly ash
replacement except for the mix with 37.5% waste
ceramic tiles replacement, which had an optimum
amount of 10% to 20% fly ash substitution.
Moreover, all combinations with 50% fly ash
substitution attained less compressive strengths
relative to mixes with 0% fly ash replacement.
The pozzolanic reaction has played a major
role in the strength development of the modified
mixes considering that both waste materials used,
ceramic tiles and fly ash, possessed pozzolanic
properties.
The k-Nearest Neighbor (k-NN) algorithm
provided a model that can predict based on the
intuitive assumption that objects close in distance
are potentially similar. The Observed Compressive
Strength vs. Observed Compressive Strength by
the k-Nearest Neighbor (k-NN) algorithm was
compared, and their plots are near the equality line,
thus, acceptable.
To further improve the conduct of the study, it
is recommended to provide superplasticizers or
other additives in the mixes in order to address the
high absorption rate of ceramic tiles that lead to
poor workability. In this way, it could result to
better applicability in the construction industry. In
addition, the fly ash and waste ceramic tiles
replacement could be limited to 0% to 30% and be
extended to 0% to 100% substitutions.
5. ACKNOWLEDGEMENT
The authors express their gratitude to De La
Salle University for providing the avenue to
conduct the laboratory experiments under this
study. They also acknowledge the Engineering
Research and Development for Technology
(E.R.D.T.) The program under the Department of
Science and Technology (D.O.S.T.) of the
Republic of the Philippines for the financial
support it has contributed to the study.
6. REFERENCES
[1] Fransman, E. J. (2009). Fly ash: An ancient
material for modern day challenges. Retrieved
from http://www.map-
abcdf.com.ph/documents/presentations/Agribu
siness/Food%20Processing%20and%20Manuf
acturing/P-20090901.pdf.
[2] Zimbili, O., Salim, W., & Ndambuki, M.
(2014). A review on the usage of ceramic
wastes in concrete production. International
Journal of Civil, Environmental, Structural,
Construction and Architectural Engineering,
91-95.
[3] Dungca, J., Galupino, J., Alday, J., Baretto,
M., Bauzon, M., & Tolentino, A. (2018).
 International Journal of GEOMATE, Aug., 2018 Vol.15, Issue 48, pp.169-174
Hydraulic conductivity characteristics of road
base materials blended with fly ash and
bottom ash. International Journal of
GEOMATE.
[4] Dungca, J., Galupino, J., Sy, C., & Chiu, A.
(2018). Linear optimization of soil mixes in
the design of vertical cut-off walls.
International Journal of GEOMATE.
[5] Galupino, J., & Dungca, J. (2015).
Permeability Characteristics of Soil-Fly Ash
Cut-Off Wall. ARPN Journal of Engineering
and Applied Sciences, 6440-6447.
[6] Galupino, J., & Dungca, J. (2016). Modeling
of Permeability Characteristics of Soil-Fly
Ash-Bentonite Cut-off Wall using Response
Surface Methodology. International Journal of
GEOMATE, 2018-2024.
[7] Galupino, J., & Dungca, J. (2017). Artificial
neural network permeability modeling of soil
blended with fly ash. International Journal of
GEOMATE.
[8] Certeza, L., & Tangquerido, N. (2010).
Effects of ceramic tiles wastes as a partial
substitute for aggregates on the compressive
strength of ordinary Portland cement and
concrete. Manila: De La Salle University.
[9] Nikoo, M., Kerachian, R., & Alizadeh, M.
(2017). A fuzzy KNN-based model for
significant wave height prediction in large
lakes. Oceanologia.
[10] American Concrete Institute. (1991). Standard
practice for selecting proportions for normal,
heavyweight and mass concrete, ACI 211.1.
[11] American Society for Testing and Materials.
(2018). Specific gravity and absorption of
coarse aggregates. American Society for
Testing and Materials, ASTM C-127.
[12] American Society for Testing and Materials.
(2018). Specific gravity and absorption of fine
aggregates. American Society for Testing and
Materials, ASTM C-128.
[13] American Society for Testing and Materials.
174
(2018). Unit weight and voids in aggregate.
[14] Ceballos, L., Fullante, S., & Mendoza, R.
(2016). The effect of partial substitution of
ceramic tile wastes as coarse aggregates in the
compressive strength of concrete. Manila: De
La Salle University.
[15] Cheng, S., Lizardo, L., Ong, J., & Rieta, A.
(2016). Compressive strength analysis of
concrete with waste ceramic tiles as a
substitute to aggregates. Manila: De La Salle
University.
[16] American Society for Testing and Materials.
(2015). Standard test method for chemical
analysis of hydraulic cement, ASTM C-114.
[17] Senthamarai, R., & Manoharan, P. (2005).
Concrete with ceramic waste aggregate. Cem
Concr Compos, 2413–2419.
[18] American Society for Testing and Materials.
(2016). Standard specification for concrete
aggregates, ASTM C-33.
[19] Bronshtein, A. (2017, April 12). A Quick
Introduction to K-Nearest Neighbors
Algorithm. Retrieved from A Quick
Introduction to K-Nearest Neighbors
Algorithm:
https://medium.com/@adi.bronshtein/a-quick-
introduction-to-k-nearest-neighbors-
algorithm-62214cea29c7
[20] Quiros, A., Bedruz, R., Uy, A., Abad, A.,
Bandala, A., Dadios, E., & Fernando, A.
(2017). A kNN-based Approach for the
Machine Vision of Character Recognition of
License Plate Numbers. 2017 IEEE Region 10
Conference (TENCON) (pp. 1081-1086).
Malaysia: IEEE.
[21] Statsoft. (2018, February). k-Nearest
Neighbors. Retrieved from Statsoft:
http://www.statsoft.com/Textbook/k-Nearest-
Neighbors.
Copyright © Int. J. of GEOMATE. All rights reserved,
including the making of copies unless permission is
obtained from the copyright proprietors.